Working Paper 04-02 The Food Industry Center University of Minnesota Printed Copy $25.50 DIET QUALITY AND CALORIES CONSUMED: THE IMPACT OF BEING HUNGRIER, BUSIER AND EATING OUT Lisa Mancino and Jean Kinsey March 2004 Lisa Mancino, Econoic Research Service, USDA, 1800 M Street, Roo N4083, Washington, DC 20036-5831 Jean Kinsey, Professor, Departent of Applied Econoics, and Co-Director, The Food Industry Center, University of Minnesota, St. Paul, MN 55108-6040, e-ail: jkinsey@un.edu. The work was sponsored by The Food Industry Center, University of Minnesota, 317 Classroo Office Building, 1994 Buford Avenue, St. Paul, Minnesota 55108-6040, USA. The Food Industry Center is an Alfred P. Sloan Foundation Industry Study Center.
DIET QUALITY AND CALORIES CONSUMED: THE IMPACT OF BEING HUNGRIER, BUSIER AND EATING OUT Lisa Mancino and Jean Kinsey ABSTRACT While Aericans clai to be eating better and iproving their understanding of diet and health, they are getting heavier and increasing their risk of suffering fro diet related illnesses. The cause of this inconsistency is unclear. Using theoretical odels of preference reversal and econoetric epirical analysis, this study finds that the nuber of calories eaten per eal increases and the quality of the diet decreases as people wait ore than six hours to eat their next eal, work ore than fifty hours a week, and consue a larger aount of food away fro hoe. These situational factors are iportant even for consuers who have considerable knowledge about diet and health. Regardless of one s favored dietary prescription, this study shows how well an individual s intentions to eat healthfully changes with tie pressures, hunger, and food source. As people change their dietary goals based on prevailing nutritional lore, such situational factors will continue to interfere with one s long-ter health objectives. This is especially relevant in an era where obesity is a leading health issue for individuals and for the costs of health care. Any advice and action that can iprove diet quality and reduce caloric intake on a convenient basis is valuable for individuals and the overall econoy.
Working Paper 2004-02 The Food Industry Center University of Minnesota DIET QUALITY AND CALORIES CONSUMED: THE IMPACT OF BEING HUNGRIER, BUSIER AND EATING OUT Lisa Mancino and Jean Kinsey Copyright 2004 by Lisa Mancino and Jean Kinsey. All rights reserved. Readers ay ake verbati copies of this docuent for non-coercial purposes by any eans, provided that this copyright notice appears on all such copies. The analyses and views reported in this paper are those of the authors. They are not necessarily endorsed by the Departent of Applied Econoics, by The Food Industry Center, or by the University of Minnesota. The University of Minnesota is coitted to the policy that all persons shall have equal access to its progras, facilities, and eployent without regard to race, color, creed, religion, national origin, sex, age, arital status, disability, public assistance status, veteran status, or sexual orientation. For inforation on other titles in this series, write The Food Industry Center, University of Minnesota, Departent of Applied Econoics, 1994 Buford Avenue, 317 Classroo Office Building, St. Paul, MN 55108-6040, USA; phone (612) 625-7019; or E-ail tfic@apec.un.edu. Also, for ore inforation about the Center and for full text of working papers, check our World Wide Web site [http://foodindustrycenter.un.edu].
TABLE OF CONTENTS Introduction and Proble Stateent...4 Theoretical Model...7 Econoetric Model...17 Epirical Estiation...24 Results...25 Conclusion, Recoendations, and Future Research...33 Appendix A...36 Appendix B...37 Study Liitations and Further Research...37 References...39
FIGURES Figure 1: Figure 2: Effect of Increasing Hunger on Leisure Tie...14 Effect of Increasing Hunger on Food Preparation Tie...15 Figure 3: Effect of Increasing Hunger on Food Consuption with Tf * =0...16 Figure 4: Figure 5: Figure 6: Interaction aong Health Inforation, Hunger and HEI score...32 Interaction aong Health Inforation, Food Source, and HEI score...32 Interaction aong Health Inforation, Tie Constraints and HEI score...33 TABLES Table 1: Variables and Definitions...20 Table. 2: Table 3: Table 4: Table 5: Variables Used to Construct Inforation Index*Results...22 Deterinants of Diet Quality (Healthy Eating Index)...26 Deterinants of Caloric Deand...27 Relative Difference of the Effect of Inforation on Diet Quality...31
DIET QUALITY AND CALORIES CONSUMED: THE IMPACT OF BEING HUNGRIER, BUSIER AND EATING OUT Introduction and Proble Stateent Over the past twenty years, the standard public policy approach for convincing Aericans to adopt healthy lifestyles has centered on disseinating inforation about the benefits of eating well. Nuerous national capaigns have aied at educating Aericans on the iportance of a healthful diet: Consue Five a Day for fruits and vegetables; Follow the food guide pyraid; Liit the percent of your daily calories fro fat. The success of these capaigns has likely contributed to the growing nuber of shoppers who adit their grocery purchases are affected by health concerns and believe eating well is a better way to anage their health than edication (FMI 2001). Despite our increased understanding of the link between diet and health, our aggregate diets do not see to be leading to better health. Recently both acadeic journals and the popular edia have been filled with reports that describe the recent trends in obesity by using ters like epideic and crisis. To recap soe of the key, often cited nubers as of 2003, 65 percent of all Aericans are overweight and over one-third are obese. Within the past ten years, the nuber of individuals classified as obese has increased by 74 percent (CDC, 2004). 1 During this sae tie span, there has been a parallel rise in the incidence of diseases highly correlated with poor nutrition and over consuption: cancer, strokes, heart disease and diabetes. According to the Surgeon General, obesity accounts for $117 billion a year in direct and indirect econoic costs, 1 An individual is classified as obese if his or her Body Mass Index (BMI), or ratio of one s weight in kilogras to one s squared height in eters exceeds 30. An individual with a BMI between 25 and 30 is classified as overweight. 4
it is associated with 300,000 deaths each year, and will soon overtake tobacco as the ain cause of preventable deaths (Surgeon General, 2004). While Aericans clai to be eating better and iproving their understanding of diet and health, they are getting heavier and increasing their risk of suffering fro diet related illnesses. The cause of this inconsistency is unclear. It ay be that Aericans just eat too uch of everything. There ay be a clear division between the people who eat poorly and the people who eat healthfully. Alternatively, it ay be that individuals try to incorporate their beliefs about healthy eating into their food choices, but soeties forego good intentions for ore iediate gratification because of tie constraints, hunger, and a deand for convenience. A rift between long-ter objectives and short-ter desires can lead to preference reversals and tie-inconsistent choices, where individuals ake selections, perhaps under pressure or in haste, which would not have been ade with an objective perspective and longer tie horizon. They ay drive over the speed liit when late for a eeting, drink too uch wine at a party, or opt to super-size at the drive-through window. If soe food choices prove to be tie-inconsistent, then our understanding of the relationship between health knowledge and diet quality can be iproved by including factors that increase one s deand for iediate gratification. Despite the ounting evidence that shorter tie delays are correlated with inconsistent choices, traditional econoic studies of consuer behavior have relied priarily on prices, incoe, and inforation to explain observed food choices. Advances in behavioral econoic theory suggest that incorporating factors, such as the delay between alternative activities and one s sensitivity to tie-delay into consuer choice analysis will further clarify the link between intentions and actual behavior (Hoch and Loewenstein; Frederick, Loewenstein, and 5
O Donoghue; Loewenstein; Baueister; Mullainathan and Thaler; Loewenstein and Elster; Thaler). It ay be that when an individual is hungry and pressed for tie, one s short-ter deand for convenience and relief fro hunger ay supercede long-ter health objectives. Since convenient foods are often high in calories, fat, siple carbohydrates, cholesterol, and sodiu, there is an ancillary increase in the consuption of these nutrients. Not accounting for one s level of hunger and deand for convenience ay lead to a isspecification of the roles of prices, incoe, and inforation on nutrient deand. In turn, this can lead to isguided public intervention progras eant to iprove nutrition, diet, and health. A better understanding of how situational factors ipact food choices will provide additional avenues for public intervention progras aied at iproving food choices and ultiately reducing the incidence of obesity. Knowing when individuals are apt to forego health concerns can be used to suggest ways to avoid such situations. Alternatively, this knowledge can be used to suggest effective coitent echaniss that would copel individuals to ake choices that are ore haronious with their long-ter health goals. This study develops a theoretical consuer choice odel that allows one s deand for convenience to change with tie pressures and hunger. Using epirical evidence, the results show that situational factors influence food choices and that the use of nutritional inforation changes as one becoes hungrier, busier, and eats ore foods away fro hoe. The results of this analysis can be used to infor policy recoendations on effective ways to iprove diet and nutrition. Before proceeding, it should be noted that while any nutritionists continue to advocate a low fat diet where the ajority of calories are derived fro coplex carbohydrates such as vegetables and whole grains, a high protein, higher fat diet has gained popularity as an effective 6
eans for weight reduction. This debate shows that knowledge about what constitutes a healthful diet is not static. Thus this study will utilize a flexible definition of health inforation to deterine how an individual s attitudes, beliefs, perceptions, and knowledge about what constitutes a healthful diet influence food choice. And because our ipression of a healthful diet is evolving, this study will analyze an individual s consuption of total calories and overall diet quality. Finally, although the ipetus for this study is to deterine econoic factors behind obesity, the analysis will focus on individual food choices. The reason for this is twofold. First, food choices play a large role in deterining body weight. Second, although situational factors are likely to affect both body weight and food choices, the effect will be ore iediate on food choices and thus, epirically ore observable. Theoretical Model The theoretical odel in this study begins with the Becker (1965) household production odel, where individuals are assued to axiize utility, subject to their ability to produce goods and services for personal use, their budget constraint, and the constraints on their tie. To ore closely depict how individuals ake their food choices, the odel in this study also draws fro a growing literature on behavioral econoics and allows individual s choices to be effected by tie-delay, hunger, and other situational factors. Several behavioral-econoic studies that have incorporated sensitivity to delayed outcoes have eployed a quasi-hyperbolic discount function, siilar to the one below. t (1) Max U = U t + β = δ U t+ 1. T t = 1 7
By allowing individuals to be ore sensitive to tie delays that occur sooner rather than later, this relatively siple depiction captures a key feature of dynaic choice odels and allows for consuption choices to be inconsistent over tie. However, as noted by Loewenstein (1996) a liitation of this odel is that hyperbolic discounters will always exhibit ipulsive behavior when tie delays are short. An iplication of this is that the hyperbolic odel cannot account for an individual who behaves ipulsively in one situation and exercises ore control in another, even if the reward values and tie delays are the sae in both situations. For exaple, a hyperbolic discounting odel would predict that an individual would always choose the ore iediately available food alternative, regardless of his or her level of hunger. To address this shortcoing, he advocates the inclusion of visceral influences, such as hunger, thirst, and pain, into an individual s instantaneous utility function. The odel in this study incorporates the idea of visceral influences to depict how individuals food choices are affected by tie delays and situational factors. Specifically, an individual in this odel akes consuption decisions on a per-eal basis () over soe finite planning period that ends at M. Utility is derived fro food ( F ) 2, a coposite non-food ite (NF ), leisure tie (TL ) and the individual s perceived health status ( H ) (Grossan). For siplicity, the utility function is assued to be separable in all arguents. It is assued to be strictly increasing and linear in health and the coposite non-food ite. With respect to food and leisure consuption, the utility function is assued to be continuous, twice continuously differentiable, strictly increasing, strictly concave, and satisfy the Inada conditions. 2 F is the total aount of food consued at, easured in gras. 8
A vector of relevant visceral factors ( α ) experienced at the tie the individual akes his or her consuption decision influences the level of utility received. To isolate the effects of situational factors on food consuption decisions, it is assued that visceral factors only influence the utility derived fro food and leisure. Thus, increasing the level of hunger experienced at tie will increase the utility garnered fro food and leisure tie, but will not affect enjoyent derived fro health or non-food. Individuals are assued to be naïve and treat these visceral factors ( α ) as exogenous M (2) U = U ( NF, H,( F, TL ; )) + δ U ( NF, H,( F, TL ; α )) α. = 1 The utility received fro health is assued to be a strictly increasing, linear function of how uch an individual knows about health and nutrition (η): (3) U( H ) = η H. This allows people who know ore about the links between diet, nutrition, and health to perceive a greater health ipact fro a change in body weight than individuals who know little about diet and health. An individual s perceived health at tie is a function of how uch that individual weighs (w ). Weight is assued to have a positive ipact on health up to soe point W* and a negative ipact thereafter. The change in weight at tie is a strictly increasing, linear function of the difference between the nuber of calories (or nutrients) consued (K -1 ) and the nuber of calories expended (E -1 ) 3. This leads to the following health production function: (4) = h( w,k E ). H 1 1 1 + + + + + 3 For siplicity, it is assued that E is exogenous. This prevents the odel fro allowing extree exercising to balance caloric intake. 9
Individuals in this odel transfor the foods they eat, such as a haburger and French fries, into calories, fat and protein through a vector of coefficients ( ε ). It draws fro the Lancaster (1966) fraework, where ε j can be interpreted as the aount of the jth nutrient contained in ( F ). It dictates an individual s perception of how uch of a specific characteristic flows fro the foods he or she consues. Consequently, individuals in this odel anage their health through the aount of calories and nutrients that they think they have consued, not the aount they have truly consued. Because individuals frequently underestiate the fat and caloric content of foods prepared away fro hoe (Kennedy et al), this odel assues that the accuracy of this coefficient vector increases with the aount of tie spent preparing food at hoe. As such, the level of error in the vector of coefficients is assued to be strictly decreasing and convex in the aount of tie an individual spends preparing food ( Tf ). This allows individuals to ore accurately assess the nutrient content of food they prepare than food prepared by soeone else so that the perception of calories consued in period are adjusted by the aount of tie spent preparing the food at tie.. In this fraework then, the foods consued are translated into calories and nutrients in the following anner: (5) Kˆ = F ε ( Tf ). Thus, the health production function, (4), to be rewritten as: (4 ) H h( w,f (Tf ) E ) = ε, 1 1 1 1 where health is a function of last periods body weight, perceived caloric intake and energy expenditures. 10
Within each planning period, individuals decide how to divide their total available tie 4 ( TT ) between working (TY ), preparing food ( Tf ), and leisure (TL ). The aount of tie spent preparing food is assued to increase as a function of the aount of food consued( F ) 5. The per-period tie constraint is: (6) TY + Tf F + TL = TT. Within each planning period, an individual decides whether to spend his or her incoe on food or non-food. P NF is the price of non-food and P F is the price of food. With a wage rate of y, individuals face the following per-period budget constraint: (7) P NF + P F y TY. NF F Food prices typically increase with the level of pre-preparation. For exaple, the cost of a raw egg is typically around $0.10, while the cost of a hard-boiled egg is around $0.75. To account for this relationship, this odel explicitly assues that the onetary price of food decreases with the aount of tie spent in food preparation in the following way: P F = P ~ btf. F P ~ F is the onetary cost of a fully prepared, ready to eat food ite. P in is the lower liit on food prices and is the price of raw aterials 6. The price-savings fro spending an additional unit of tie preparing food is b. Using the preceding egg exaple, if it took one unit of tie to prepare a hard boiled egg, then b would be equal to $0.65. Noralizing the price of non-food to 4 Total available tie is the aount of tie in a day, less tie required for sleep. 5 For siplicity, it is assued that food production technology exhibits constant returns to scale. 6 To ensure that individuals do not sell food they produced, it is assued that 0< Tf < ( P ~ F -Pin)/b. 11
one and aking it the nueraire, the tie, and budget constraint can be cobined into the fullincoe constraint: ~ (8) NF + ( PF btf ) F = y TT y Tf F y TL. Rearranging (8), the coposite non-food ite can be expressed as a function of prices, wages, and the aount of food consued at tie : ~ (8 ) NF = y TT y Tf F y TL ( PF btf ) F. Substituting the health production function (4 ) and the full-incoe constraint (8 ) into (2), the individual s inter-teporal optiization proble becoes: (9) MAX U + M = 1 δ U = U ~ ( ytt ytf F ytl ( P btf ) F )), H,( F, TL ; α ))) (( NF, h( w, F ε ( Tf ) E ),( F, TL, α )). + + 1 + 1 + 1 F + 1 + + + Because the axiization proble is finite-diensional and the constraint set is closed and bounded, the Bolzano-Weierstrass theore guarantees that a solution to the axiization proble exists. Thus, the axiization proble is well defined. Furtherore, because the constraint set is convex, the utility function is strictly concave and satisfies the Inada conditions, the first order conditions are both necessary and sufficient to ensure that the axiization proble yields a unique solution. Optiizing (9) with respect to the variables F, TL, and Tf 12
yields the following first order conditions: (9a) (9b) (9c) U F U TL U Tf = y Tf M * h δ η w * ~ ( P = 1 + = y + U = y F TL + ( TL * + bf F w F ; α * btf + + ε ( Tf ) = 0 M = 1 * ) + U * F h δ η w ( F ) = 0 * + + w, α ) + [ 0, M ], ε + [ 0, M ], ε Tf and F * 0 [ 0, M ]. Solving for Food (F ), Leisure Tie (TL ), and Food Preparation Tie ( Tf ) and substituting the utility axiizing levels of F * and Tf * into the (5) yields the reduced for deand function for nutrients and/or calories: ˆ K ~ (10) K = D ( PF, b, y, α, η, E, w 1). Equation (10) iplies that nutrient deand will be a function of prices, price savings, wages, visceral factors (hunger), inforation, physical activity, and one s body weight. Equation (9b) shows that in equilibriu * U TL ( TL ; α ) = y. This iplies that increasing the level of hunger experienced at will increase the optial level of leisure tie because 2 U ( TL, α) TL α > 0 and thus, increasing α increases U TL. As shown in Figure 1, this suggests that an increase in α to αˆ will lead to an increase in the optial aount of TL * to TL **. 13
Figure 1: Effect of Increasing Hunger on Leisure Tie $ MC TL MB = U TL TL, αˆ ( MB = U TL ( TL, α ) ) TL * TL ** TL * Assuing Tf > 0, equation (9c) shows that optiality iplies that an individual will equate the arginal benefits of tie spent preparing food to the arginal cost. For soeone whose weight exceeds W*, (9c) can be rewritten as: M b + δ = 1 h η w + + w ε + ε Tf = y = U TL ( TL * ; α ) = U TL ( TT Ty * Tf * ; α ). Thus, the arginal cost of an extra unit of food preparation tie is the loss of tie spent in leisure. If hunger increases, the arginal cost of preparing food will also increase. As shown in Figure 2, increasing hunger fro α to αˆ will decrease the optial aount of Tf * to Tf ** 14
Figure 2: Effect of Increasing Hunger on Food Preparation Tie $ MC = U TL TL, αˆ ( ) MC = U ( TL TL, α ) M h w MB = b + δ η + + = 1 w + ε ε Tf Tf ** * Tf Tf Finally, (9a) illuinates how increasing hunger at eal will influence overall food consuption at that eal: In equilibriu, an individual will equate the arginal benefits of food consuption to the arginal cost. For soeone whose weight exceeds W*, (9a) can be rewritten as: M * * ~ * h + w + * (9a ) U F ( F, α ) = ytf + ( PF btf ) + δ η ε ( Tf ). w F If = 1 + Tf * * =0, then increasing α will lead to a higher value of U ( F F, α ) and have no effect on the right hand side of (9a ). Thus, to satisfy the first order conditions, an individual will increase F * until (9a ) holds with equality. As shown in Figure 3, increasing the level of hunger fro α to αˆ will increase the optial level of food consuption fro F * to F * *.This result is iportant. If an individual decides that the optial aount of food preparation tie at eal is zero, then he or she will likely purchase food that is already prepared. This suggests that the 15
effect of hunger on the aount of food chosen will be greater for food prepared away fro hoe than food prepared at hoe. Figure 3: Effect of Increasing Hunger on Food Consuption with Tf * =0 $ ~ MC = P F M h η + δ + w + = 1 w + F ε(0) MB = U F F, αˆ ( MB = U F ( F, α ) ) F * F ** F The effects of hunger on optial food consuption is abiguous when increasing Tf * >0. Fro Figure 2, α will decrease Tf *. Buying food that is ore prepared ay lead to a decrease in future health and an increase in the current onetary cost. However, reducing Tf * will decrease the arginal tie cost of preparing food. As α increases, the perceived opportunity * cost of preparing foods, ( U ( TL ; α )), increases as well. Thus, it is possible for the total TL change in the arginal cost of increased consuption caused by an increase in α to be negative, zero, or positive. Nevertheless, as long as the positive change in the total arginal benefit exceeds the total increase in arginal cost, the optial level of food consuption will increase with hunger. How uch an individual knows about health and nutrition η, the price 16
savings fro preparing one s own food (b), and the opportunity cost of one s tie (y) all ipact the likelihood that increasing hunger α will increase F *. Overall, the equilibriu conditions suggest that hunger will have a significant and positive ipact on nutrient deand when: 1. An individual consues food prepared away fro hoe; 2. The price savings fro food preparation are low; 3. The opportunity cost of tie is high; and 4. An individual has less knowledge about health and nutrition. Econoetric Model With observations on i individuals who consued eals, the following functional for is used to estiate per-eal nutrient and caloric deand: ( 12) K = β ' X + e, where K is a 1 J 7 vector of observations on actual per-eal nutrient consuption, β is a 1 n vector of coefficients, X is an n J atrix of explanatory variables, and e is a 1 J error vector. However, changing the unit of observation on nutrient and caloric consuption to a pereal basis transfors what would norally be a cross-sectional data set into a panel data set. Using OLS to estiate nutrient deand with such data ay yield inefficient paraeter estiates because individuals reporting ore than two eating occasions on a single day would provide at least two observations for estiation. Another econoetric issue stes fro the fact that hunger can be influenced by the aount of food consued at the previous eating occasion. Not accounting for this relationship could also lead to inefficient paraeter estiates because the 7 J is equal to the total nuber of observations, or the nuber of individuals, ultiplied by their nuber of eating occasions. 17
disturbance ter fro the past eating occasion ay be correlated with the error ter fro the current eating occasion. To circuvent these econoetric issues, this study estiates the ean relationship between the dependent and explanatory variables for a single individual. This allows (12) to be rewritten as follows: ( 13) K = β ' X + e, i i i where i denotes the individual, K i is the average per-eal level of consuption, X i is a the per-eal average value of the n explanatory variables, and e i is the per-eal average error. With this functional for, OLS estiation techniques should yield both consistent and efficient estiates. Another benefit of averaging per-eal nutrient deand over the two days of intake is that it provides a better idea of how a series of choices affects overall nutrient intake and how people balance their caloric and nutrient consuption over an entire day. A third benefit of averaging per-eal factors and outcoes is that it allows for the use of the Healthy Eating Index (HEI) as a dependent variable. This index suarizes an individual s diet quality over an entire day. The coponents that coprise this index are an individual s servings of fruits and vegetables, carbohydrates, fat, protein, cholesterol, and overall variety in their diet (Bowan et al). Another econoetric issue is that several of the right hand side variables, naely health inforation, hunger, and food source, are arguably endogenous. As such, these variables will be correlated with disturbances and ay yield biased and inconsistent paraeter estiates. The standard econoetric ethod of correcting for probles of easureent error and endogeneity is to use soe type of instruental variables (IV) estiator. This requires that one find variables that are highly correlated with the endogenous explanatory variables and not correlated with the 18
error ter in the equation being estiated. However, the low correlation aong variables eans that IV estiators ay still be biased and inefficient (Park and Davis). For that reason, this study eploys both IV and OLS estiates. This study uses STATA 7.0 and specifies the survey s priary sapling unit, the level of stratification, and sapling weight to account for the data s coplex survey design. The resulting paraeter estiates will be ore efficient than those that would have resulted fro siply using either OLS or IV estiators. This study estiates the effects of inforation and situational factors on the deand for overall calories and the deand for overall diet quality, easured by an individual s HEI score. When estiating the deand for calories, the dependent variable is the two day average of the total calories an individual consued at each eating occasion divided by that individual s total recoended daily allowance (RDA). When estiating overall diet quality, the dependent variable is the individual s average HEI score for the two days of intake. Suary statistics are reported in Table 1. The theoretical odel developed in this study suggests that per-eal nutrient deand will be a function of food prices, an individual s wage rate, body weight, caloric expenditures, inforation about health and nutrition, per-eal situational factors that affect one s sensitivity to tie delay, and the aount of tie spent preparing the eal. A liitation of this data set is a lack of inforation on both food prices or food expenditures. However, when individuals intake choices are ade within a short tie frae it is standard practice to assue that the odicu of variation in prices across households can be captured by inforation on the household s regional location (Variya et al, 1995, 1996). Thus, geographic location and whether or not an individual lives in an urban, suburban, or rural area are included to represent inor fluctuations in food 19
Table 1: Variables and Definitions Category Variable Definition Mean Std. Dev Dependent Variable Dcals HEI Calories at eal as a percent of RDA Individuals HEI score, 2 day intake average 0.288 62.91 0.115 11.77 Wage Rate: Incoe Size Total household incoe in $1,000 Nuber of ebers in household 34.88 2.586 26.37 1.464 School Level of schooling: 1 if less than high school 2 if high school or GED 3 if soe college 0.170 0.339 0.169 0.375 0.473 0.374 4 if at least four years of college 0.229 0.420 Food Prices and Expenditures Body Weight and Caloric Expenditures Deand Shifters Situational Factors and Sensitivity to tie delay Food Source: Inforation Midwest South West Northeast Urban Suburban Rural BMI Feale Activity Active job BFPL TV Age Vegetarian Soker White Black Hispanic Interval Breakfatst0 Brekfast1 Breakfast2 Snack Meal Free Captive Cheap Social Planned Inforation Beliefs Perceptions Awareness 1 if Midwest 1 if South 1 if West 1 if Northeast 1 if central city 1 if suburb 1 if rural Body weight (kgs)/ height 2 (eters 1 if individual is feale; 0 otherwise Stated activity level: 1 if less than 3 ties pre onth 2 if 1-4 ties per week 3 if at least 5 ties per week 1 if individual has a job that is physically deanding; 0 otherwise 1 if pregnant, breast feeding or lactating Average hour of t.v. watching per day Age of eal-planner in years 1 if vegetarian 1 if soker 1 if White 1 if Black 1 if Hispanic Tie elapsed between eals 1 if no breakfast on day 1 or day 2 1 if only one day with breakfast 1 if breakfast on both day1 and day2 1 if previous eating occasion was a snack 1 if previous eating occasion was a eal. 1 if eal cae fro soeone else 1 if eal cae fro cafeteria, dining center 1 if food cae fro fast food restaurant, pizza place, vending achine 1 if eal cae fro a sit down restaurant or bar 1 if eal cae fro a grocery store 0.252 0.355 0.203 0.191 0.296 0.437 0.267 28.09 0.496 0.442 0.290 0.265 0.224 0.009 2.667 50.88 0.030 0.257 0.776 0.115 0.081 4.104 0.058 0.156 0.786 0.336 0.677 0.076 0.022 0.083 0.064 0.713 See Table 5.2 5.993 4.619 7.022 7.605 0.434 0.479 0.402 0.392 0.457 0.496 0.442 11.82 0.500 0.497 0.451 0.441 0.417 0.095 2.176 17.19 0.171 0.437 0.416 0.318 0.272 1.422 0.234 0.363 0.410 0.179 0.182 0.137 0.079 0.133 0.120 0.098 1.668 1.378 1.296 1.390 20
prices and expenditures. The household incoe for a given individual, the size of that household, and an individual s level of education are included to provide data on an individual s wage rate. An individual s BMI and gender are included to provide inforation about an individual s body weight and caloric expenditures. The rationale for including these variables is that BMI should be highly correlated with weight, and all else equal, woen tend to weigh less than en. An individual s reported level of physical activity, whether or not the individual has a physically deanding job (Kuchler and Lin), and whether or not an individual is pregnant, lactating or breastfeeding are all factors that will increase ones caloric expenditures. Conversely, the nuber of hours of TV watched per day and an individual s age should be negatively related to one s caloric expenditures. An individual s ethnicity, whether or not an individual is a vegetarian, and whether or not the individual currently sokes cigarettes are included as additional explanatory variables in the per-eal nutrient and caloric deand functions. It is hypothesized that these variables ay act as deand shifters. Nicotine is reported to be an appetite suppressant, thus sokers will likely eat less at each eal. Due to the nature of their diets, vegetarians are likely to consue less fat and protein at each eal. Because individuals fro different ethnic backgrounds ay have very diverse dietary habits, ethnic differences ay also cause significant variation in per-eal caloric and nutrient consuption. Responses fro the Diet Health Knowledge Survey (DHKS) (Appendix A) are used to create an index to easure knowledge about health and nutrition for each individual. Typically, such an index is constructed by suing the nuber of questions an individual answers correctly about the links between diet and health. However, as the arketing axio suggests, perception is reality; what soeone perceives to be true is likely to be a better predictor of 21
behavior than siply whether or not soeone believes what experts aintain as true. Because of this, health inforation is grouped into four general categories: inforation, beliefs, perceptions, and iportance (Table 2). Justification for this is based on a theory of huan behavior developed for arketing that links beliefs and attitudes to observed behavior (Fishbein and Manfredo). Using this fraework illuinates how different aspects of inforation are used when aking food choices. For exaple, although a consuer ay be fully aware of the links between being overweight and health probles, if she does not think it is iportant, she will be less likely to act on this inforation. Table. 2: Variables Used to Construct Inforation Index* Objective Inforation Subjective Inforation Knowledge Beliefs Perceptions Iportance 5.993 4.622 7.022 7.603 1.668 1.378 1.296 1.390 Agreeent with 'soe people are Perception of caloric intake born to be fat..not uch you can do..' Nuber of correct answers on health questions Whether or not an individual is on any low calorie, low fat, low sodiu, high fiber or diabetic diet. Nuber of correct servings of grains, fruits, vegetables, etc. the individual is able to identify Nuber of diseases associated with being overweight, eating excessive aounts of calories, fat, sodiu and cholesterol. Whether or not respondent is involved with soe aspect of eal planning Iportance of aintaining a healthy weight, eating lots of fiber, eating lots of fruit and vegetables, and liiting intake of fat, sodiu, and cholesterol. Agreeent with 'What you eat can ake a difference in chance of getting a disease' Perception of own body weight, health, diet quality and nutrient intake Agreeent with '.. so any recoendations.. it's hard to know what to believe' Use of nutrition labels *The variable's ean and standard deviation are reported below each variable's nae. 22
Answers to questions that for the inforation index have definitive right and wrong answers, whereas answers to belief questions are ore subjective 8. An exaple of a question used to for the inforation index is whether an individual was able to identify the recoended daily servings of vegetables. The total nuber of diseases a respondent attributed to over consuption of fat, sodiu, cholesterol, excessive calories, and obesity is used to construct the beliefs index. The perceptions index is constructed by coparing how respondents ranked their own diet quality to how their actual diet was ranked using the USDA s Healthy Eating Index (Bowan et al; Variya et al, 2001, Mancino and Kinsey). Questions that asked respondents to rank the iportance they placed on aintaining a healthy weight, liiting saturated fat, eating fiber, and liiting cholesterol are used to construct the iportance index. In order to isolate the effects of situational factors such as hunger and sensitivity to tie delay, this study analyzes nutrient intake at each eating occasion. Although the CSFII does not explicitly ask individuals how hungry they were at each eating occasion, it does provide inforation on the tie elapsed between eating occasions, the individual s classification of each eating occasion as either a snack or a eal, and whether the individual reported eating breakfast on either or both days of the recall. These variables are used to proxy an individual s average level of hunger at each eal. Finally, the CSFII does not collect inforation on the aount of tie an individual spends preparing eals. Therefore, where the individual purchased or received the eal is used to proxy the aount of tie he or she spent preparing that eal. When an individual reported ore than one food source for a single eal, the food source that provided the ost calories was considered the source for that eal. There are five types of places the individual could have 8 A detailed description of the questions used to construct each coponent of the inforation index can be obtained 23
procured his or her eal, fro soeone else; a cafeteria, or day care center; a fast food restaurant, pizza place or vending achine; a sit-down restaurant; or a food store, such a grocery store, convenience store or superarket. Epirical Estiation This study uses three different econoetric odels to estiate per eal caloric and nutrient deand. The first odel uses OLS to estiate the following average per-eal deand functions: Model 1: OLS K D i β + β ' y + β ' p + β w + β ε + β η + β α + β FS + e, where = o 1 i 2 i 3 i ' 4 i 5 i 6 i 7 D K i is the average per-eal deand for calories, and overall diet quality (HEI) 9. These dependent variables are assued to be linear functions of an individual s incoe variables (y i ), variables that affect food prices and expenditures (p i ), factors that affect an individual s weight and caloric expenditures (w i ), other deographic characteristics (ε i ), an individual s knowledge, beliefs, perceptions, and attitudes about diet and health (η i ), the average per-eal visceral factors ( α ), and the proportion of eals that cae fro a fast-food restaurant, a full-service i restaurant, a grocery store, a cafeteria, or soeone else ( FS ). The theoretical odel developed in this study suggests that the variable Food Source should be considered endogenous. Thus, the second odel in this study uses whether or not an individual received food staps or participated in the WIC progra to instruent the portion of food prepared at hoe. The rationale behind this is that these progras liit the aount of oney spent on foods prepared away fro hoe. For exaple, food staps can be used to buy i i i fro the authors. 9 Observations are averaged over both days of intake. 24
ground-beef and haburger buns, but they cannot be used to buy a haburger at a restaurant, deli, or fast-food place. Using these two variables as instruents requires a change in the epirical odel eployed in Model 1 because there are insufficient instruents to partition food sources beyond food at hoe and food away fro hoe. For this reason, Model 2 has only one variable for (FS * i). This variable easures the share of eating occasions that consisted of foods prepared at hoe. This study uses two-staged least squares for the IV estiation. In the first stage, the variable hoe is regressed on the instruental variables plus all other exogenous variables included in Model 1. Within Model 2, the deand functions are estiated twice, once using an IV estiator and once using an OLS estiator. This is done to provide ore consistency between the odels with and without instruents. Statistical tests suggest that there is no systeatic difference between the two estiators, and thus OLS estiates will be both consistent and efficient. Subsequent description of the overall estiation results will focus on the finding fro the OLS regressions. Results Results fro the OLS regressions are suarized in Tables 3 and 4. 10 When interpreting the results, it is helpful to know that the intercept ter represents a white ale who went to college, lives in an urban area in the Northeastern United States, has a sedentary job, exercises oderately (was not pregnant, breastfeeding or lactating) and ate breakfast on both days of the dietary recall. The estiated OLS odels explained about 30 percent of the total variation in an individual s two-day average HEI score (Table 3) and nearly 20 percent of the variation in the average per-eal energy (Table 4). 10 Detailed results of the IV estiates can be found in Mancino, Lisa Dissertation or on request fro the authors. 25
Table 3: Deterinants of Diet Quality (Healthy Eating Index) Dependent Variable: HEI Estiated Beta Standard Error Intercept 58.193 1.978 ** Incoe 0.0180 0.0070 ** Hhsize -0.4569 0.1287 ** Soe High School -1.3869 0.8465 High School -0.9632 0.5080 * 4+ years of College 1.3097 0.5762 ** Midwest 0.2540 0.5971 South -1.1048 0.8266 West 0.2178 0.6978 Suburb -0.3501 0.4964 Rural -2.3626 0.7329 ** BMI -0.0377 0.0142 ** Feale 0.1549 0.3207 Very Little Exercise -0.9276 0.4983 * Very Active -0.3094 0.3918 Active Job -0.3841 0.4577 BFPL 6.7365 2.1054 ** TV -0.2095 0.0965 ** Age 0.0738 0.0112 ** Vegetarian 0.6716 1.4553 Soker -3.8763 0.4757 ** Black -2.2166 0.8043 ** Hispanic 2.1192 0.6910 ** Inforation 0.4290 0.1281 ** Beliefs 0.6075 0.1755 ** Perceptions 0.7707 0.1800 ** Iportance 0.3878 0.1415 ** Interval -1.7540 0.6236 ** Interval2 0.0526 0.0538 No Breakfast -4.7971 0.9757 ** Breakfast on One Day -3.2197 0.4686 ** Meal 1.0760 1.3825 Other eo -1.2529 2.8519 Hoe NA NA Free -2.2210 1.5190 Captive -0.5162 2.7958 Social -7.2076 1.1865 ** Cheap -7.9262 1.4478 ** Other -3.8103 1.8320 ** R-Squared 0.286 F(37,7) 32.420 *Significant at the 5% level **Significant at the 10% level 26
Table 4: Deterinants of Caloric Deand Dependent Variable: Calories Estiated Beta Standard Error Intercept 0.2423 0.0233 ** Incoe 0.0000 0.0001 Hhsize -0.0031 0.0016 ** Soe High School -0.0106 0.0080 High School -0.0045 0.0070 4+ years of College -0.0134 0.0058 ** Midwest 0.0103 0.0093 South -0.0089 0.0094 West 0.0031 0.0068 Suburb -0.0046 0.0057 Rural 0.0005 0.0058 BMI 0.0000 0.0002 Feale -0.0270 0.0040 ** Very Little Exercise 0.0018 0.0054 Very Active 0.0117 0.0074 Active Job 0.0022 0.0068 BFPL -0.0066 0.0113 TV 0.0053 0.0022 ** Age -0.0006 0.0002 ** Vegetarian -0.0107 0.0122 Soker 0.0021 0.0065 Black 0.0303 0.0138 ** Hispanic 0.0019 0.0075 Inforation 0.0019 0.0014 Beliefs -0.0006 0.0015 Perceptions -0.0082 0.0023 ** Iportance -0.0044 0.0017 ** Interval 0.0254 0.0069 ** Interval2-0.0011 0.0006 ** No Breakfast 0.0025 0.0113 Breakfast on One Day 0.0072 0.0079 Meal 0.0929 0.0164 ** Other eo 0.0239 0.0318 Hoe NA NA Free 0.0805 0.0233 ** Captive 0.0391 0.0248 Social 0.0998 0.0172 ** Cheap 0.0806 0.0378 ** Other 0.0173 0.0184 R-Squared 0.207 F( 37, 7) 11.860 *Significant at the 5% level **Significant at the 10% level 27
Situational Factors: Analysis on variables that gauge an individuals level of hunger at each eating occasion supports the hypothesis that such situational factors do have a significant ipact on actual food choices and diet quality. In all OLS estiates, increasing the interval between eating occasions is associated with a significant decrease in HEI scores and caloric consuption. The effect of hunger on food choices is estiated to increase caloric consuption up to soe point then significantly taper off. Evaluated at the saple eans, a ten percent decrease in the interval between eals is estiated to increase one s HEI score by.94% and reduce per-eal energy over 2.5%. Stated another way, decreasing the length of tie between eals by 24 inutes is correlated with a 0.6 point increase in the HEI score and 15 point reduction in calories consued at each eal. Although this sees rather sall, it should be noted that, for an individual who consues 3 eals a day, decreasing the interval between eals would reduce caloric intake by 45 calories a day. Over a year, this would result in a about a five pound reduction in total body weight, all else being equal. Individuals who reported eating breakfast on both days of intake score significantly higher on the HEI than individuals who skipped breakfast on at least one day of the recall. Eating breakfast on both days is also estiated to significantly decrease the total nuber of calories consued at each eating occasion. Health and Diet Inforation: Increasing the accuracy of an individual s perceptions about weight and diet quality significantly increases one s HEI score and significantly decreases pereal caloric intake. The level of iportance placed on aintaining a healthy diet also significantly influences overall diet quality. Individuals who report placing ore iportance on diet and health are estiated to have a significantly higher quality diet and consue significantly fewer calories at each eal. Increasing the nuber of diseases an individual associates with 28
consuing nutrients like fat and cholesterol is estiated to significantly increase an individual s per HEI score. Increasing health inforation, easured as objective knowledge about health and nutrition, significantly increases an individual s HEI score. Evaluated at the saple eans, a ten percent increase in the accuracy of one s perception about diet quality increases the HEI score over 0.80% and decreases his average per-eal caloric intake by over 1.8%. This translates to a little over a half point increase in his HEI score and reduces the calories consued at each eal by ten calories. Increasing overall health knowledge by ten percent increases the HEI score by 1.62%, about one point, and decreases average pereal energy intake by 1.17%, roughly seven calories. Food Source: Analysis on where an individual procured food for a specific eating occasion supports the hypothesis that the source of food ipacts one s food choices and diet quality. Increasing the share of eals obtained fro fast food (cheap) or full service restaurants (social) significantly decreases an individual s HEI scores and significantly increases per-eal consuption of calories copared to eating at hoe. Increasing the share of eating occasions prepared by soeone outside the hoe, such as a snack tray or a eal prepared by soeone else (free), is estiated to significantly increase per eal caloric intake. Household Incoe and Household Size: There is a positive relationship between incoe and diet quality. This is congruous with the theoretical assuption that diet quality is a noral good. Individuals fro larger households are significantly ore likely to consue fewer calories per eal and have a lower HEI score. Because incoe is often correlated with education, it is not surprising that effect of schooling was estiated to be siilar to the effect of incoe. Copared to an individual who had soe college, individuals with four or ore years of college are significantly ore likely to have higher HEI scores and consue fewer calories at each eal. 29
Price/Geographic Location: Individuals who live in rural areas score significantly lower on the HEI than individuals who lived in urban areas. There is no significant difference in calories and HEI by region of the country. Weight and Physical Activity: An individual s reported BMI was estiated to be negatively correlated with his or her HEI score. Feales consue significantly fewer calories at each eal. Individuals who reported very little physical activity have significantly lower HEI scores than individuals who exercise ore regularly. Woen who were breastfeeding, lactating, or pregnant (BFPL) score higher on the HEI. Increasing the aount of television watched per day is found to have a significantly positive effect on per-eal calorie consuption and a significantly negative effect on an individual s dietary quality. Additional Deand Shifters: Older individuals consue significantly fewer calories and have a significantly higher quality diet. Sokers have lower HEI scores. Black individuals have significantly lower HEI scores than whites, while Hispanic respondents are estiated to score significantly higher than whites. Blacks also consue significantly ore calories at each eal than white respondents. Interaction of Inforation and Situational Factors: Finally, this study tests the hypothesis that the role of inforation on food choice changes as one becoes hungrier, busier, and consues ore foods away fro hoe. For siplicity, the four health knowledge variables are aggregated into a single inforation index. Fro this, OLS estiates are used to gauge the joint effect of the interval between eals and inforation, the portion of food consued away fro hoe and inforation, and the nuber of hours worked in a week and inforation. Table 5 suarizes the results of this analysis by reporting each interaction ter s relative agnitude, its estiated coefficient, and its estiated p-value in parentheses. The first row of the second 30
colun of Table 5 should be interpreted to ean that the effect of inforation on overall diet quality significantly decreases as an individual goes longer between eals. Overall, the results of this analysis on the interaction ter suggests that the strength of the relationship between health inforation and overall diet quality wanes as an individual consues ore eals away fro hoe, goes longer between eals, and works ore hours in a given week. Table 5: Relative Difference of the Effect of Inforation on Diet Quality Relative Effect of Inforation on Overall Diet Quality (HEI) when an individual Goes Longer Between Meals Eats More Meals At Hoe Works More Hours in a Week Saller** -.1468 (.006) Larger**.9005 (.005) Saller** -.0030 (.000) **Estiated to be significant at a.05% level of significance Figures 4 through 6 show these relationships graphically. The dashed line in Figure 4 represents how an individual s HEI score will change with increasing levels of health inforation with no interaction ter. The other four lines represent how an HEI score will respond to increasing levels of inforation when an individual goes two, four, six or eight hours between eals, using the saple eans in conjunction with the estiated coefficients fro a regression that includes an interaction ter between health inforation and the interval between eals. It shows that the HEI score rises with health inforation, but when the interval is six hours or ore, the HEI falls below the ean effect (dashed line). As people get hungrier, less attention is given to healthy diets even at high levels of diet and health inforation. Siilarly, Figure 5 shows that an individual s HEI score is lower when eating ore eals away fro hoe. Working 50 and 60 hours in a given week also pushes the HEI below the ean (Figure 6). 31